Clock jitter is the deviation of clock edges from their ideal timing. Because that deviation eats into the time available to capture data and degrades high-speed links and data converters, jitter must be kept to a minimum in a high-speed system.
High-speed signaling is very sensitive to jitter. As signals toggle faster, the transmitter and receiver face tighter limits. In many high-speed data applications the clock edge must fall within a narrow window to capture data correctly. The more jitter in a system, the more often the edge falls outside that window, and how often that happens shows up as the system's bit error rate (BER).
What jitter is, in plain words
Picture a perfect clock as a row of edges spaced exactly one period apart. A real clock's edges wander a little early or late around those ideal positions. That wander is jitter. National Instruments puts it simply: jitter is the difference between when an event should happen and when it actually happens (NI, Digital Timing: Clock Signals, Jitter, Hysteresis, and Eye Diagrams).
Jitter is small, usually picoseconds or less, but a fast clock has very little time to spare. At 1 GHz a whole clock period is only 1,000 ps, and at 10 Gb/s one bit lasts just 100 ps.
Where it comes from
- Random sources: thermal and other device noise in oscillators, PLLs and clock buffers. This part is unbounded in principle and is described statistically.
- Deterministic sources: power supply noise, crosstalk from neighboring signals, simultaneous switching outputs, and data-dependent effects on a channel. This part is bounded and often repeats.
Why it matters
- Digital timing: inside a chip, jitter shortens the effective clock period, so less time is left for logic between flip-flops.
- Serial links: in SerDes, Ethernet and PCIe receivers, jitter closes the data eye and raises the bit error rate.
- Data converters: an ADC samples at the clock edge, so a mistimed edge samples the wrong voltage. Texas Instruments shows that with fast input signals, clock jitter becomes the main limit on an ADC's signal-to-noise ratio (TI, Clock jitter analyzed in the time domain, Part 1).
How engineers keep it down
- Use a clean reference oscillator and a PLL with suitable loop bandwidth.
- Give PLLs and clock buffers their own quiet, well-decoupled supplies.
- Route clocks away from aggressors and keep the clock tree short and balanced.
- Budget jitter explicitly in timing analysis and link design rather than hoping margin covers it.
Deep dive: the numbers and details
This part holds the definitions, equations and worked examples. Skip it if the overview is all you need.
Ways to measure jitter
| Measure | What it compares | Typical use |
|---|---|---|
| Period jitter | Each clock period against the ideal period | Setup timing inside a chip |
| Cycle-to-cycle jitter | Each period against the one before it | PLL and oscillator data sheets |
| Time interval error (TIE) | Each edge against its ideal position | Serial links, long-term accumulation |
| Phase jitter | Phase noise integrated over a frequency band | Reference clocks for transceivers and ADCs |
Each figure can be quoted as an rms value or as a peak-to-peak value. Always check which one a data sheet gives; the two differ by a large factor.
Random and deterministic jitter
Total jitter (TJ) depends on how rare an error you are willing to accept. The dual-Dirac model splits jitter into a deterministic part (DJ, peak-to-peak) and a random, Gaussian part (RJ, rms), then estimates TJ at a chosen BER:
TJ(BER) = DJ + alpha(BER) x RJ alpha = about 14 at BER = 1e-12
The value of about 14 for BER 10-12 is the one used by Teledyne LeCroy in its description of the model (Teledyne LeCroy, Understanding SDA II jitter calculation methods).
Worked example (illustrative values): a 10 Gb/s link has a unit interval of 100 ps. With DJ = 10 ps and RJ = 1 ps rms, TJ at 10-12 is about 10 + 14 x 1 = 24 ps, or 24% of the bit time. Note how the 1 ps of random jitter contributes more than the 10 ps of deterministic jitter at this error rate.
Jitter in static timing analysis
For a path between two flip-flops on the same clock, the setup check with jitter is:
T_clk - t_jitter >= t_clk-to-q + t_logic + t_setup (+ clock skew terms)
In STA tools the jitter usually goes into the clock uncertainty value. Example: a 1 GHz clock (1,000 ps) with 30 ps of period jitter leaves 970 ps for the path, so 3% of the cycle is gone before any logic is placed. Hold checks compare launch and capture on the same edge, so cycle jitter largely cancels there; teams usually apply a smaller hold uncertainty.
Jitter and ADC signal-to-noise ratio
For a sine wave input of frequency fIN sampled with rms jitter tJ, the jitter-limited SNR is:
SNR_jitter (dBc) = -20 x log10(2 x pi x f_IN x t_J)
TI's example: 400 fs of jitter at a 100 MHz input limits SNR to about 72 dBc, whatever the converter's resolution (TI SLYT379). Doubling the input frequency costs 6 dB. Independent random jitter sources, such as the clock and the ADC's own aperture jitter, add as the square root of the sum of squares:
t_J,total = sqrt(t_J,clock^2 + t_J,aperture^2)
Common mistakes
- Adding rms jitter values linearly instead of root-sum-square, which overstates random jitter.
- Mixing rms and peak-to-peak numbers from different data sheets.
- Quoting phase jitter without the integration band; the same oscillator gives different numbers over different bands.
- Forgetting that a PLL can pass, filter or amplify reference jitter depending on its loop bandwidth.
- Using a total-jitter figure without stating the BER it applies to.
Further reading
- NI white paper: Digital Timing, clock signals, jitter, hysteresis and eye diagrams
- Texas Instruments, Clock jitter analyzed in the time domain, Part 1 (Analog Applications Journal, 3Q 2010)
- Teledyne LeCroy, Understanding SDA II jitter calculation methods (dual-Dirac model)


Your comments will be moderated before it appears here.